01.10 Power First: The Constraint That Rules Everything

Lesson

Picture a developer holding everything: a thousand flat acres under option outside a growing metro, equity committed, a contractor ready to mobilize. The project dies in a single letter from the utility. No capacity on this feeder until 2031. The land is still there. The money is still there. The crew is still available. None of it matters.

That failure has a name: the binding constraint. Every project has many requirements, but only one of them decides what happens: the one that runs out first, takes the longest, or has no substitute. Sort the inputs in this story by that test:

Land: substitutable. Another parcel exists.
Capital: substitutable. Another lender exists.
Equipment and labor: substitutable. Another vendor, another crew.
Grid power: not substitutable. One site gets one answer from the utility about when the megawatts arrive, and that answer rules the project more surely than any contract you will sign.

The first question in any deal conversation is "do you have power?" It has a precise meaning: does the utility say the megawatts are real, contracted, and dated? The industry sorts every announcement into two piles:

Real megawatts: the utility has committed capacity, with a date.
Press-release megawatts: a campus announced without a utility answer. That is a rendering, not a project.

The veterans stopped believing land deals years ago. They believe the utility's answer.

One honest qualification: the binding constraint moves. Power rules first, not forever. Once the megawatts are secured, something else becomes binding, and each lesson in this curriculum names the next one. The sequence teaches one skill: find the binding constraint at every stage, starting with the one that kills projects before they start.

Think in cities, not buildings. When the plans say 100 MW, do not picture a building. Picture a small city, because the grid sees it exactly that way: a city appearing on a feeder.

100 MW draws like roughly 80,000 homes of continuous load.
Modern AI campuses run into the hundreds of megawatts each.
The IEA projects data centers worldwide will use slightly more electricity by 2030 than the whole of Japan does today.
The frontier is now gigawatt scale. xAI's Colossus cluster in Memphis reached roughly 1 GW of AI compute by mid-2026, with full site draw over 1.2 GW at peak. It began with an 8 MW grid connection. The rest came from gas turbines parked on site and, later, a dedicated 1.2 GW gas plant. Read that again: a gigawatt of demand behind an 8 MW wire.
For calibration, the entire city of Chicago uses about 30.2 TWh a year, an average of roughly 3.45 GW of continuous load. A single 1 GW campus is nearly a third of Chicago.

At this scale the constraint sharpens. You are no longer asking for capacity on a feeder. You are asking for new transmission lines and new generation, which are multi-year regional planning decisions. The largest projects now start with the power answer and work backward to everything else.

One equation runs underneath everything that follows:

\text{PUE} = \text{Utility MW} \div \text{IT MW}

\text{Utility MW} = \text{IT MW} \times \text{PUE}

Every megawatt of servers drags its overhead multiple of cooling and support load behind it. Memorize this equation. The curriculum assumes it everywhere. At gigawatt scale it reads the same: 1,000 MW (1 GW) of IT at PUE 1.4 is 1,400 MW (1.4 GW) from the grid.

PUE has a history, and the history has stalled.

2007: The Green Grid introduces PUE. The industry average is 2.5.
2014: The easy fixes have driven the average to 1.65.
2025: Uptime Institute's survey puts the global average at 1.54, flat for six years.

The average hides two industries. Hyperscale operators run 1.1 to 1.2 at their best sites. Smaller enterprise facilities cluster around 1.6 to 1.8. Germany now writes the target into law: 1.5 by 2027, 1.3 by 2030.

Before you trust a PUE number, remember two things:

PUE measures the facility, not the IT. It says nothing about water use or server efficiency, and a half-empty hall scores worse than a full one.
PUE is annual and weather-dependent. A quoted PUE without a climate and a measurement period is marketing.

Reference

A data center is a power business that happens to house computers.
- The product being bought, built, and sold is not floor space. It is megawatts.
- Every decision downstream (the site, the equipment, the schedule, the deal) traces back to how much power, where it comes from, and when it arrives.

The vocabulary

Capacity: the maximum power a system can deliver.
A nameplate rating, not a measurement of what is actually drawn.
Taught in 01.10 Power First: The Constraint That Rules Everything
kWkilowatt: a unit of electric power equal to one thousand watts.See also MW, GWTaught in 01.10 Power First: The Constraint That Rules Everything
GWgigawatt: a unit of electric power equal to one billion watts.
The unit for the largest AI campuses and regional load planning.
See also kW, MWTaught in 01.10 Power First: The Constraint That Rules Everything
IT MW: The amount of electrical power delivered directly to servers, storage drives, and networking gear inside the data hall.
It measures the actual productive power consumed by the computing hardware doing the work. It excludes facility overhead.
Also called Critical Load, Critical IT Load, IT Equipment EnergyTaught in 01.10 Power First: The Constraint That Rules Everything
Load: total amount of raw electrical power being drawn by a system
Moves with IT utilization; the quantity that matters operationally.
See also IT MW, Utility MWTaught in 01.10 Power First: The Constraint That Rules Everything
MWmegawatt: a unit of electric power equal to one million watts.
The unit for the largest AI campuses and regional load planning.
See also kW, GWTaught in 01.10 Power First: The Constraint That Rules Everything
PUEPower Usage Effectiveness: a standard metric used to measure the energy efficiency of a data center.
Calculated by dividing the Utility MW by the IT MW
$\text{PUE} = \text{Utility MW} \div \text{IT MW}$ $\text{Utility MW} = \text{IT MW} \times \text{PUE}$
The overhead multiple of the compute load.
Ideal is 1.0, which means 100% of the utility power directly fuels the IT equipment (servers, storage, and networking). No real facility gets there.
Rule of thumb: 1.5 is average, under 1.3 is excellent, above 1.8 is legacy, 1.1-1.2 is hyperscale
See also IT MW, Utility MWTaught in 01.10 Power First: The Constraint That Rules Everything
Utility MW: The total amount of raw electrical power drawn from the public power grid to run the entire data center campus, including cooling and support loads (lighting, security, office, losses).
Always larger than IT MW.
Also called Total Facility Energy, Grid CapacityTaught in 01.10 Power First: The Constraint That Rules Everything

The scale

A 100 MW campus draws like a small city: roughly 80,000 US homes of continuous load.
Modern hyperscale campuses run into the hundreds of megawatts.
Gigawatt-scale AI campuses are now being built. 1 GW is 1,000 MW: roughly 800,000 US homes of continuous load.
xAI's Colossus cluster in Memphis reached ~1 GW of AI compute by mid-2026, drawing over 1.2 GW at peak. It began on an 8 MW grid connection.
For calibration: Chicago uses about 30.2 TWh a year, roughly 3.45 GW of average continuous load. A 1 GW campus is nearly a third of Chicago.
When a procurement plan says "100 MW," it means a power quantity first. Square footage, rack count, server specs are all downstream of the megawatts.

Why power sets site feasibility, schedule, and cost

Site feasibility: without the megawatts on the needed timeline, there is no project. Everything else can be in place and it does not matter.
Schedule: the binding timeline is the power timeline: when the megawatts arrive. Other phases can overlap or compress; this one sets the date.
Cost: land, equipment, and electricity all price off the megawatts.
Not land, not capital: land without power is a field with a fence, and money cannot shorten the one timeline it does not control.

The mental model

Follow the megawatts and you find the project.
When a schedule slips, find the power constraint. When a budget breaks, find what the megawatts cost. When a site dies, ask what the utility said.

Case study: The Letter from Dominion

Fictitious-entity notice: Meridian Data Centers, its development director, and the project figures below (150 MW, PUE 1.4, 80 acres, the 18-month clock, the tenant LOI) are illustrative inventions. Dominion Energy, Loudoun County, and the July 2022 events are public record, sourced at the end.

Situation

It is August 2026. You are the development director for Meridian Data Centers, planning a 150 MW IT-load campus in eastern Loudoun County, Virginia ("Data Center Alley," the largest concentration of data centers on earth, with some 26 million square feet already operating in the county).

You hold everything a project needs. The land: 80 acres under contract, price locked. The money: equity committed, debt term sheet signed. The customer: a letter of intent from a hyperscale tenant for the full first phase, with delivery expected in 2029. The county wants you there: data centers supply roughly a third of Loudoun's property tax revenue, enough to cover the county government's entire operating budget.

Then, in July 2026, Dominion Energy, the utility serving the region, tells its data center customers that new power connections in eastern Loudoun will be delayed for years. The problem is not generation. There is enough electricity. The problem is delivery: the high-voltage transmission lines and substations into this pocket of the county are full, and a project needing a new line cannot connect until that line is built. Dominion establishes a "constrained area." Full power for new projects there will not be available until 2030 at the earliest, a date later pushed to mid-2032.

Your land contract requires you to close or walk away within 18 months. Your tenant's LOI expects 2029 delivery and goes soft after that. Your investors are asking whether to break ground on schedule.

The decision

Recommend one: build, hold, or move.

How to solve it

Name the binding constraint in one sentence. What is the one input with no substitute?
Run the lesson's tools: convert the IT load to utility draw with the PUE equation, and sanity-check the scale against the small-city test.
Test each option against the utility's date, not against your preferences, your sunk costs, or your tenant's wishes. An option that requires power before the utility's date is not an option.
Commit: build, hold, or move. State the single fact your recommendation hinges on, and what would change your mind.

A strong answer fits on one page and never confuses a building with a project.

Exhibits

A, the load math. Planned IT load: 150 MW. Design PUE: 1.4. Utility draw: $150 \times 1.4 = 210$ MW, roughly the continuous draw of 170,000 US homes.
B, the dates. Land decision: within 18 months (by early 2028). Tenant expects 2029 delivery. Dominion's earliest full-power date: 2030, later revised to June 2032.
C, the options. (1) Build on schedule: close the land, start construction, deliver the building on time without the power. (2) Hold: extend the land option if possible, wait for the transmission build. (3) Move: shift the project to a nearby market with available power, and start over on land and permitting.

What not to use

Stay inside this lesson's tools: the binding constraint, the MW vocabulary, the PUE equation, the scale test, follow-the-megawatts. Don't reach for the internal anatomy of the power chain, equipment lead-time tables, interconnection-queue mechanics, redundancy architectures, formal site-selection methods, cost models, or on-site generation. Those come later in your learning, and this case doesn't need them.

Sources and fictitious entities

Modeled on public record: Dominion's July 2022 customer notifications and "constrained area" as reported by The Register (July 29, 2022) and Data Center Frontier; Loudoun County tax and square-footage figures via The Register. The 2026 setting and the 2032 horizon are projected forward for the exercise. Illustrative inventions: Meridian Data Centers, the development director persona, and all project figures. This case uses only public record and clearly labeled hypothetical material.

Branden's Answer

Recommendation: Immediately move the project to a market with contractually guaranteed power availability.

1. The Binding Constraint

The problem is that the hyperscaler tenant expects delivery in 2029, while Dominion cannot provide the required 210 MW of utility power until later, probably MUCH later. Even though everything else is in place, nothing can overcome the lack of electricity delivered by the utility company.

2. Metrics and Model

This project is a 150 MW IT load site with a respectable projected PUE of 1.4 (far short of the hyperscale gold standard of 1.1 or 1.2) . At such a ratio, that 150 MW floor will draw 210 MW from the utility.

\text{IT MW} \times \text{PUE} = \text{Utility MW}

\text{150 MW} \times \text{1.4} = \text{210 MW}

That is roughly equivalent to the continuous electricity demand of 170,000 US homes. The utility's transmission constraint makes power delivery the primary constraint on the critical path, not land acquisition, financing, or construction readiness.

3. Test the three options

Option	Assessment	Decision
Build	If construction proceeds, but the project cannot meet its 2029 delivery commitment without power. The building alone does not fulfill the tenant's requirement.	Reject
Hold	Waiting for power risks losing the tenant and tying up capital. The revised June 2032 power date falls well beyond the expected 2029 delivery and the early-2028 land decision deadline.	Reject unless the land option and tenant commitment can be renegotiated on acceptable terms
Move	Relocating creates new land and permitting work, but a market with verified power availability offers a credible path to delivery.	Recommend

4. Commitment and decision trigger

Move the project. First confirm the alternative site can deliver the full 210 MW on the tenant's schedule. Available land and preliminary utility assurances are not a basis for commitment.

The decision rests on one fact: Dominion's earliest stated date for full power arrives after the project's commercial deadline.

Caveat: Consider holding the current site if:

Dominion produces a credible, documented power-delivery date that fits the tenant's requirements, and
the land option and tenant commitment can be extended without unacceptable economic exposure.

5. Conclusion

This project lives or dies on megawatts of power. We know that Dominion cannot deliver them in time, so the site as held has no path forward. Building without power produces a stranded asset. Holding burns the tenant and the capital. Moving preserves the commercial opportunity.

The lesson: follow the power, not the sunk costs. Lock the power timeline first, then commit the land, the capital, and the construction schedule around it.